Mechanical thinking drove the industrial revolution, and data thinking detonated the intelligent revolution. This is the worst world, and it is also the best world. At the moment, we are in the midst of an intelligent revolution. Cloud computing, big data, and artificial intelligence technologies are contributing to the intelligentization of various industries.
Driven by smart technology, what changes will our social activities produce? In which direction will our future city develop? New technologies represented by AI are promoting the further development of smart cities.
AI + security application bottlenecks and difficulties in the construction of smart cities
With the rapid development of artificial intelligence applications, bottlenecks and difficulties in AI+ security applications have become more prominent
Bottleneck 1: "Algorithmic Dividend" is about to disappear
As the most well-known algorithm competition in the field of artificial intelligence, ImageNet is currently the world's largest database for image recognition. This project was established by computer scientists at Stanford in the United States, simulating the recognition system of humans. After the development from 2010 to 2015, the face recognition algorithm has developed from a shallow neural network to a complex deep neural network algorithm, from the original (2012) 8-layer neural network to the 152-layer neural network in 2015.
The recognition rate has also increased rapidly, from the 28.2% error rate of the shallow network to 3.57%. From 2010 to 2016, the recognition rate of various algorithms of artificial intelligence has been rising. The core reasons are: 1. The rapid increase in hardware computing power, especially the wide application of GPU, allows the neural network model to be larger; 2. The neural network algorithm itself also After evolution and self-support, the algorithm efficiency and recognition rate have been further improved. Today, with the establishment of an ultra-deep neural network model with 1,000 layers at every turn, the neural network algorithm has reached data saturation, and the recognition rate of speech and image has entered the bottleneck area. If there is no breakthrough in the algorithm level, the future recognition algorithm The efficiency improvement will rapidly diminish marginally.
At the same time, with the efforts of leading artificial intelligence research institutions represented by Google Research and Microsoft Research in the open source of algorithms, software engineers can quickly carry out "secondary development" through artificial intelligence development kits provided by Google, Microsoft and other companies. ", to realize the application of artificial intelligence in a new field. With the rapid decline of the threshold for artificial intelligence research and development, startups under the banner of "artificial intelligence" have sprung up around the world, and have gained popularity in the capital market, and competition in the artificial intelligence market has become increasingly fierce. Entrepreneurs in the field of artificial intelligence in China have stated that the dominant position of artificial intelligence algorithms can only be maintained for 6 months at most, and industry hegemons such as Google and Baidu can only maintain the lead in a single algorithm for 6 months. "Algorithmic dividends" are about to disappear. How to use algorithms to find the core pain points of users and put them into application is the magic weapon for artificial intelligence companies to survive.
Bottleneck 2: Three major barriers are increasingly emerging
The underlying technology of artificial intelligence has completed the original accumulation, no matter from the various recognition algorithms of perceptual intelligence (weak intelligence) to surpass the human senses, or cognitive intelligence (strong intelligence and semi-strong intelligence) in AlphaGo after defeating Li Shishi, it was aliased in early 2017. AlphaGo, an upgraded version of Master, has successively defeated new and old chess sages such as Nie Weiping and Ke Jie on the Go website. At this stage, artificial intelligence has polished its core deep neural network algorithm extremely sharp. Countless entrepreneurs and companies have also quickly formed artificial intelligence teams and are trying to get a share of the artificial intelligence era. First, Google hired GeoffreyHinton, the founder of deep neural networks, Facebook hired YannLeCun, and Baidu hired AndrewNg. In addition, DeepMind was established and acquired by Google, and companies such as SenseTime and Megvii Technology began to start businesses and achieved global leadership in their respective fields. However, these companies are facing the same problem-how to industrialize their technology and get market recognition.
These events show that the Internet leaders are beginning to realize that algorithms will become less and less scarce, and that the marginal improvement of algorithms will become smaller and smaller. These leading companies have adjusted their AI strategies through personnel changes and used their AI products as products. Strategic Direction. At a time when the input-output ratio of artificial intelligence algorithm research and development becomes no longer economical, and the performance of the algorithm itself has reached a mature level, the "algorithm dividend" in the AI ​​field is about to disappear. We believe that data scarcity, productization capabilities, and channels Resources will build the core barriers to the industrialization of artificial intelligence.
Barrier 1: Data scarcity
The establishment of neural network algorithm model requires a large amount of data to be tuned and improved. In a sense, the larger the amount of training data, the better the data quality, and the higher the accuracy of the algorithm obtained by training, so the data in a specific field Will become scarce resources and barriers for industry access. From the value point of view, the three types of data in security, finance, and medical care have both industry thresholds and strong data monopoly. For example, in the security field, if you want to perform face comparison and identity authentication, the algorithm company must be authorized by the relevant government department to extract ID card information. Furthermore, if an artificial intelligence diagnostic image recognition company needs a large amount of raw data of medical images, a large number of doctors or technicians who understand medical expertise are required to mark the preliminary data, so as to ensure the reliability of data quality.
Barrier 2: Productization ability
The application of artificial intelligence involves more engineering productization capabilities, which will also become one of the core competitiveness of artificial intelligence companies to survive. For example, in the face recognition system that pays for facial recognition, the algorithm company succeeded in adding "in vivo verification" (blinking, turning head, etc., proving that the user is a living person, not a photo) process through continuous trial and error. Compared with overseas companies, Chinese companies have stronger engineering and order-based R&D capabilities. On the one hand, productization capabilities can complement some of the weak disadvantages of "late" companies in algorithm efficiency, and on the other hand, it can greatly improve the functions and efficiency in vertical scenarios. Technically speaking, almost all artificial intelligence startups have relevant algorithm capabilities, but when integrated into specific scenarios, engineering and productization capabilities will become a magic weapon for companies to bypass "roadblocks."
Barrier three: channel advantage
We believe that artificial intelligence technology will be the first to be applied to ToG and ToB end applications, while ToC end applications are lagging behind. The core reason is that the scenes in the ToG and ToB fields are often simple and the business repeatability is high. Therefore, artificial intelligence is expected to rapidly improve business efficiency and reduce manual investment. Since ToC applications need to be customized and adjusted according to individual users, in the early days of artificial intelligence, the experience is often poor when the data volume is insufficient, and it is difficult to promote on a large scale. In the ToG and ToB fields, artificial intelligence technology has begun to land large-scale orders in many fields such as security, education, and finance. The landing of the initial order will greatly reduce the business model threshold and form a "snowball effect": large-scale applications will also accumulate a large amount of data, which in turn will rapidly improve the accuracy of the model. The first-mover advantage brought by this "snowball effect" of channel advantage will become the core advantage of artificial intelligence companies.
Generally speaking, China’s video surveillance industry has gone through four stages: 1. The initial stage, traditional analog surveillance, the domestic independent intellectual property rights are backward, and the security system users are limited to government departments; 2. The development stage, digital surveillance, security users increase, surveillance Scale expansion, image digital storage, resolution entered the SD era; 3. Improved stage, high-definition monitoring, market capacity continues to increase, video monitoring system and user business system integration; 4. Intelligent stage, gradually forming integrated data transmission, video , Control-in-one intelligent security integrated management platform. China is currently in the stage of improving high-definition surveillance. At this stage, the market capacity continues to increase, and the video surveillance system is integrated with the user's business system. With the upgrading of video surveillance technology, the intelligence of the security industry is also rapidly coming. At the same time, the improvement of hardware equipment performance also provides an important guarantee for artificial intelligence applications. The deep neural network algorithm used in the image recognition function is constructed using a large amount of training set data for simulation and tuning, which means that the operation and application of artificial intelligence image recognition technology cannot be separated from the support of a large amount of data, which provides the model. The richer the image information, the higher the pixels, and the higher the definition, the more helpful it is to improve the accuracy of image recognition. As an important development direction of the industry, the security industry in the future must be a high-tech industry. The combination of AI + security will provide the impetus for the industry to enter the intelligent stage.
The application prospects and trends of AI + security in the construction of smart cities
In the construction of smart cities, it can be seen from smart transportation that in the future, artificial intelligence police robots will replace traffic police, realizing all-round monitoring of road traffic safety, all-weather patrols, and three-dimensional supervision. At present, improving the road traffic safety prevention and control system is a major scientific and technological construction project of the national public security traffic management department. The highway traffic safety prevention and control system realizes timely monitoring, discovery, evidence collection, transmission, processing, feedback, and correction of vehicle traffic, traffic violations, and road hazards on highways, further improving the strength and level of highway management and control, and further enhancing the scientific and scientific nature of service management. Targeted, timely detection and correction of various traffic violations, significantly improved road traffic order, and effectively curbed major traffic accidents.
The core technologies involved in the highway traffic safety prevention and control system are traffic behavior monitoring, traffic safety research and judgment, traffic risk warning, and traffic law enforcement. These technologies are now integrated with artificial intelligence. Realize the "visible" road traffic operating status, the "understandable" of the vehicle trajectory, the "capable" of key violations, the "eliminable" safety hazards, the "fast response" of road cooperation linkage, and the "service of traffic information application" Targets such as "excellent" are inseparable from artificial intelligence technology.
The future application of artificial intelligence is to slowly complete the things that are done manually in traffic management through systems and equipment. The degree of intelligence is getting higher and higher. The primary and intermediate decision-making will gradually be replaced by artificial intelligence decision-making. Intelligent transportation products and technologies are nothing more than to achieve this goal. As to whether the application of artificial intelligence in the transportation field can develop rationally depends on the company itself, but it is certain that demand cannot promote rational development.
The future of intelligent transportation must be the urban traffic brain to connect and manage all the intelligent facilities of urban traffic. All traffic data is collected into a "brain", with thousands of experts assisting decision-making solutions, and the management mode of each department All can be matched with the system, and all decisions made with cross-departmental linkage can be executed smoothly. It should be reminded that, like many other technologies, artificial intelligence cannot solve all problems, especially some non-technical problems. Wang Xiaojing, chief scientist of the Institute of Highway Science of the Ministry of Transport, believes that it is necessary to avoid excessive "deification" of intelligent transportation technology. Application effects, forming a scientific transportation development structure through scientific planning and system design is still the key task of urban transportation development, and the function of intelligent transportation technology is to support service upgrades.
Concluding remarks
Artificial intelligence is the future of the security field. On the road to the future, there are still many obstacles and difficulties that need to be overcome and overcome, but the overall trend is optimistic. We firmly believe that there are only autonomous, personalized, and constantly evolving artificial Intelligent brains can solve the increasing demands in the security field, become experts and assistants for the majority of users, improve the intelligence level of the entire security field, and promote the upgrading of the security industry.
Draw-wire sensors of the wire sensor series measure with high linearity across the entire measuring range and are used for distance and position measurements of 100mm up to 20,000mm. Draw-wire sensors from LANDER are ideal for integration and subsequent assembly in serial OEM applications, e.g., in medical devices, lifts, conveyors and automotive engineering.
Linear Encoder,Digital Linear Encoder,Draw Wire Sensor,1500Mm Linear Encoder
Jilin Lander Intelligent Technology Co., Ltd , https://www.jllandertech.com